NRV Pinch: Some Data on Data Centers
A data center can be a very good tax-base project and supplemental to the overall infrastructure while being a fairly mediocre employment project. It can also be a giant, resource-sucking pollutant. For the New River Valley, the question should not be “Are data centers good or bad?” It should be whether this particular facility, on this particular site, pays enough into the community to justify the electricity, land, water, noise, heat and infrastructure it consumes. Virginia's own JLARC reached essentially that conclusion: the industry has genuine economic value, but most of it arrives during construction, while the operating workforce is surprisingly small for the enormous buildings and capital involved.
Hey, What’s Going On?
As recently as June, data centers were permitted by right in the General Industrial I-2 district. Town staff brought the issue to council specifically because a by-right project would leave Christiansburg considerably less leverage over setbacks, noise, water, generators, traffic and other conditions. Council has since scheduled a September 8 public hearing on changing the ordinance so a data center would require a Conditional Use Permit.
Then there’s the site at Pulaski County at the New River Valley Commerce Park, near Dublin. Early this year, officials confirmed that VFRIFA already had a developer under contract but no end user. Fresh August reporting says a developer has now acquired approximately 475 acres under a $750 million development agreement that could result in a data center, but an actual operating customer still has not been identified. In other words, this has advanced beyond idle speculation, but it is still not the same thing as Microsoft, Google, Meta or another hyperscaler announcing “we are building X megawatts here.”
The Commerce Park is, in fairness, one of the more rational places in the NRV to put such a thing if one is going to be built. It is already a 1,000-acre industrial park, outside the 100-year floodplain, with environmental, wetlands, geotechnical and cultural-resource studies already performed. It has 138-kV Appalachian Power service, fiber, and guaranteed capacities of 1 million gallons per day each for potable water and wastewater. It was explicitly marketed for data center/information-technology development before the current controversy.
So, Christiansburg should decide what rules apply before somebody comes knocking; Pulaski/VFRIFA now needs to decide whether the actual deal is good enough.
What do we get out of it?
Some say jobs… Large skilled-trade workforce; grading, concrete, electrical, HVAC, pipefitting, trucking, contracting, technical, electrical, HVAC, security and maintenance work meals/lodging and suppliers. A typical building takes 12–18 months and can peak around 1,500 construction workers. A large campus can keep construction going for five years or longer. Kind of nothing, especially if outsourced.
Well-paid technical, electrical, HVAC, security and maintenance work. JLARC's typical 250,000-square-foot center employs only about 50 full-time workers, roughly half company employees and half full-time contractors. Average pay across these jobs was around $100,000 in JLARC's statewide study, but the employment density is tiny compared with many manufacturing or distribution uses. Totally nothing in the grand scheme of this area.
Others say tax money! Enormous taxable physical plant and equipment must mean a lot of tax revenue! A big stimulus! The headline investment number is deceptive as a measure of local economic activity. JLARC estimates construction is only about 20% of total data-center capital investment; 68% is IT/mechanical equipment, much of it purchased outside Virginia. A "$750 million investment" emphatically does not mean $750 million circulates through Pulaski businesses.
In Virginia's mature markets, data-center taxes ranged from less than 1% to 31% of total locality revenue. But the payoff depends heavily on the local tax rate and whatever abatements or incentives are granted. Give away too much tax to win the project and the strongest argument for hosting it starts eating itself.
VFRIFA was created specifically so participating localities could share development costs and revenues. That makes this particular site more defensible regionally than a conventional privately held industrial parcel whose fiscal benefit stops at a county line.
It’s a tax farm of computers. The converse of a solar farm. That is not necessarily an insult. I see the value of some facilities, like ones that support infrastructure.
And a facility that throws off millions in dependable tax revenue while requiring relatively few new school seats, police calls or other population-driven services can be a very useful municipal asset. Pulaski administrator Jonathan Sweet has made essentially that argument: because the center would not demand thousands of workers from an already constrained regional labor pool, it could complement rather than cannibalize Volvo, advanced manufacturing and other employers.
But that logic only holds if the tax money stays substantial and the public doesn't quietly wind up financing the infrastructure needed to produce it.
Electricity is the biggest systemic concern
Water gets more headlines, but power is probably the biggest regional issue.
Virginia electricity consumption had been broadly flat for years before data-center growth changed the curve. JLARC's modeling found unconstrained state demand could double within ten years, principally because of data centers, requiring immense amounts of additional generation and transmission infrastructure.
And that is no longer merely a Northern Virginia problem. Appalachian Power told Virginia regulators this summer that it has 36 prospective large-load customers, primarily data centers, requesting roughly 25,000 MW combined. Appalachian Power's entire recent system peak was only about 7,200 MW. Most of those proposals will not necessarily materialize, but the scale demonstrates what is now knocking on Southwest Virginia's door.
There is a legitimate upside. Southwest Virginia historically lost enough industrial demand that parts of Appalachian Power's network have capacity looking for customers. A large, steady load can use existing infrastructure more efficiently and potentially spread fixed costs across more electricity sales. Regulators have also begun making large-load customers accept longer contracts and minimum-payment obligations to reduce the chance that a speculative data center causes billions in infrastructure construction and then vanishes. Appalachian Power's new large-load tariffs include initial contractual commitments of at least 14 years. Virginia SCC
But that protection should not be mistaken for magic. A gigawatt-scale customer can require substations, high-voltage transmission, generation and regional grid upgrades well beyond its own fence. The NRV should not accept a deal where residential and small-business ratepayers finance a server company's electrical expansion.
And until the Pulaski developer identifies the end user and announces the proposed megawatt load, nobody can responsibly calculate the project's real regional cost.
That missing MW number is arguably more important than the $750 million headline.
Water: potentially minor, potentially enormous
This one requires resisting bullshit from both camps.
It is false that every modern data center sucks millions of gallons of water every day. It is equally false that modern data centers inherently use negligible water.
JLARC obtained actual Virginia utility data and found enormous variation. Most individual buildings used about as much as, or less than, an average large office building—about 6.7 million gallons annually. But 11 buildings exceeded 50 million gallons in a year, and one used 243 million gallons. Cooling technology makes the difference: dry cooling uses little or no water but more electricity; evaporative cooling can dramatically reduce electrical cooling loads while consuming much more water. JLARC
That tradeoff is especially important here because the entire NRV Commerce Park currently advertises a guaranteed potable-water supply of 1 MGD. NRV Commerce Park
For perspective, the 243-million-gallon Virginia example averages roughly two-thirds of that entire guaranteed daily capacity by itself.
That does not mean the proposed Pulaski center will use that amount. We don't know its cooling system. And that is exactly the problem.
Before approval, the public should get:
annual potable-water consumption;
maximum summer-day consumption;
cooling technology and expected water-use effectiveness;
how much water is recirculated;
whether reclaimed/non-potable water can replace drinking water;
drought-stage restrictions;
expected wastewater volume and chemistry.
Those are engineering numbers, not ideological opinions.
Heat pollution is becoming a real siting issue
This concern has moved from theoretical to measurable.
A peer-reviewed 2026 Arizona State University study measured four air-cooled Phoenix-area data centers and found their condenser exhaust itself was 14–25°F hotter than ambient air. Downwind neighborhoods averaged about 1.3–1.6°F warmer, with observed increases up to 4°F, detectable as far as roughly one-third of a mile in some conditions. ASU News
That finding should not simply be pasted onto Christiansburg or Dublin. Phoenix's dry climate, topography, wind patterns and cooling architecture are nothing like an Appalachian valley. The magnitude here could be substantially different.
But the physical principle isn't controversial: almost every watt a data center consumes eventually becomes heat, and that heat has to go somewhere.
For a hyperscale air-cooled facility, Christiansburg or VFRIFA should require site-specific thermal-plume modeling, particularly if homes, streams, parks or sensitive ecosystems lie downwind.
Trees and setbacks are not merely aesthetic landscaping at that point. They become infrastructure.
Noise is probably the most immediate quality-of-life threat
This is where good siting makes an enormous difference.
Cooling fans, chillers, transformers and electrical equipment can produce a constant low-frequency hum. JLARC reviewed troublesome Virginia sites with measurements roughly between 40 and 59 dBA—not generally dangerous to hearing and often below conventional noise limits, but residents describe the constant character of the sound as disruptive to sleep and well-being. About 10% of operational Virginia sites examined appeared to have generated significant noise complaints. JLARC
That makes the Commerce Park much more defensible than dropping one beside a Christiansburg subdivision. It already exists as an industrial site and was intentionally designed to separate intensive development from ordinary residential land use. WVTF Public Radio
Still, “industrial” should not become shorthand for make whatever fucking noise you want.
A permit should specify sound limits at the property boundary, specifically account for low-frequency noise, establish generator-testing hours and require independent preconstruction modeling plus post-opening measurements.
Land and ecological disruption
A data center's direct land impacts are mostly familiar industrial-development impacts: tree clearing, grading, soil compaction, impervious roofs and pavement, stream and wetland disturbance, stormwater runoff, habitat fragmentation, transmission corridors and access-road construction. Virginia already regulates many of those effects, though private undeveloped land itself has comparatively limited protection. JLARC
Again, this makes the Commerce Park the comparatively sensible site. It is already designated industrial, outside the 100-year floodplain and has undergone wetlands and environmental studies.
The real local ecological question is scale.
A reported 475 acres is almost half of a 1,000-acre regional industrial park. Even if only part is ultimately developed, dedicating that much prepared industrial land to a low-employment use carries an opportunity cost. The park is also marketed for advanced manufacturing, aerospace, pharmaceuticals, distribution and other uses. NRV Commerce Park
That acreage could potentially support several companies employing far more people.
Which means the data center needs to win not merely against empty land, but against the best plausible alternative use for the same scarce prepared industrial acreage.
The realistic balance sheet
Cornelius's read of the evidence comes out roughly like this:
During construction, the economic case is legitimately strong. Hundreds to roughly 1,500 workers per building at peak, high-value electrical and mechanical work, site preparation, trucking and several years of possible campus buildout can inject real money into the region. But it is temporary, and the huge investment headline exaggerates what stays local because most server and equipment spending leaves Virginia.
After construction, the employment case becomes weak but the fiscal case can become very strong. A building the size of a Walmart Supercenter might have about 50 people running it. What the locality gets instead is an enormous taxable building full of enormously expensive equipment. JLARC
That means a data center should never be sold to the NRV principally as “jobs.”
If it gets built, the sales pitch should read:
“This is a low-employment, high-capital industrial taxpayer, and in exchange for letting it consume an extraordinary amount of infrastructure we intend to extract enough revenue to improve life for everybody else.”
That is an intellectually coherent deal.
But then the locality has to actually extract the revenue.
What the NRV should require before saying yes
A project this resource-intensive should get approval only after the developer provides, in binding and independently reviewable form:
CUP/special-permit review, never automatic by-right development, with the end user publicly identified before final approval.
Exact maximum MW demand, phased energization schedule and generator count—not merely square footage or investment dollars.
Developer responsibility for customer-specific substations, grid connections and other attributable upgrades, backed by long-term commitments so ratepayers do not inherit a stranded billion-dollar extension if the AI boom goes sideways.
Exact annual and peak-day water use, cooling technology, drought plan and preference for closed-loop, dry, hybrid or reclaimed-water systems where practical.
Binding low-frequency noise limits, generous setbacks, vegetated buffers and restricted nighttime generator testing.
Modern low-emission backup generation and transparent reporting of generator testing and emissions.
A site-specific thermal-plume study, because waste heat is now a demonstrated local environmental effect rather than merely a theoretical one. ASU News
Stormwater overdesign, stream/wetland buffers, meaningful preserved green space and an enforceable electronic-waste recycling program.
A 20- to 30-year public fiscal model showing gross local taxes minus incentives, roads, utility upgrades, public-service costs and other subsidies.
Performance requirements and clawbacks tied to actual capital investment, jobs and wages—not “anticipated economic impact.”
A serious opportunity-cost comparison per acre against advanced manufacturing, aerospace, pharmaceutical and other employers the Commerce Park was also built to attract.
And some defined share of the net new public revenue should go straight into the boring shit that makes the NRV better to live in: drainage, sidewalks, transit, schools, water systems, fire/EMS, parks and attainable housing.
Because that last point gets to the whole argument.
The data center's strongest defense is if the data center pays for the sidewalk. Its worst-case scenario is if the sidewalk winds up paying for the data center.
Bottom line
Christiansburg: requiring a CUP before any project arrives is the correct approach. There is no reason to surrender negotiating power to an unidentified future hyperscaler. The town should write the rules while nobody has millions of dollars riding on changing them. Christiansburg Official Website
NRV Commerce Park: the site itself is about as rational as the region could offer—existing industrial zoning, strong fiber, major electric infrastructure, water/sewer capacity and regional revenue-sharing. But the project cannot yet be judged conclusively because the most important facts remain undisclosed: end user, electrical demand, cooling design, water use, total buildout and final incentive/tax agreement.NRV Commerce Park
So at this stage, the evidence supports neither “DATA CENTERS WILL SAVE SOUTHWEST VIRGINIA” nor “THE SERVER BUILDING WILL POISON THE NEW RIVER.” The Arsenal is already doing that.
Like most things, it’s nuanced.
Yes, perhaps—but only if it pays its own freight, water, electricity, pollution controls and infrastructure costs, produces enough durable public revenue to compensate for the land it occupies, and is kept the fuck away from people's bedrooms.
